PRATIK PRIYANSHU

M.Sc. thesis project. HNEP (Hybrid Network Evaluation Protocol, v3.0) is a reproducible benchmarking framework for quantum-classical hybrid learning that combines graded surrogation, structural interventions, and convergent validity analysis across 7 model families and 4 molecular datasets. It introduces the Quantum Contribution Taxonomy — the first two-dimensional classification of quantum roles in hybrid models — and shows that single-method QML evaluations can produce systematically incomplete or contradictory conclusions.

ExoVeil is a self-supervised, prediction-based transit detection system for exoplanet science. It learns each star’s quiescent photometric behaviour and treats transits as departures from that baseline, enabling single-transit detection — a regime in which phase-folding classifiers (ExoMiner, AstroNet, RAVEN) score 0% by construction. Released as an open-source package (pip install exoveil) with pretrained weights and a candidate catalogue.

Proposed cMSCI (calibrated Multimodal Semantic Coherence Index), a novel geometric metric for tri-modal semantic coherence evaluation that integrates Gramian volume geometry, contrastive calibration, and uncertainty-aware adaptive weighting (including training-free Matryoshka scale-consistency estimation). Instantiated on both dual-space (CLIP + CLAP, 512-d) and unified-space (Gemini Embedding 2, 3072-d) backbones, with an ensemble of uncorrelated errors outperforming either individually.

A multi-agent environmental monitoring system that predicts Harmful Algal Blooms (HABs) across German lakes by fusing satellite imagery, water quality sensors, weather data, and visual analysis through autonomous AI agents communicating via Google's Agent-to-Agent (A2A) protocol.

A research framework for building trustworthy legal AI systems, evaluated on 251,038 real German court decisions. It investigates how retrieval strategy — from dense embedding search to citation graph-augmented hybrid retrieval — affects the quality, faithfulness, and fairness of AI-assisted legal decision support.

An LLM-powered multi-agent system for analyzing traffic accident liability under German traffic law (StVO). It orchestrates nine specialized agents through LangGraph — from YOLOv8 scene perception to CAN bus telemetry parsing to RAG-augmented legal reasoning — and compares three structurally distinct pipeline variants against 30 hand-authored ground-truth scenarios.

A research-grade, trust-aware Retrieval-Augmented Generation (RAG) system that integrates domain gating, hybrid retrieval, evidence clustering, contradiction detection, and confidence calibration to minimize hallucinations in high-stakes environments.

A modular ADAS perception and safety stack integrating camera, LiDAR, and radar fusion with interaction-aware prediction, explainable AI, safety diagnostics, and scenario validation, built to mirror German OEM architecture principles.